What Is It Like To Be a House?

Exploring a Few Deceptively Simple Questions About Housing Supply

For all that the economics blogosphere is big into talking about housing supply and demand, I’ve only rarely actually seen anyone provide any good data on actual housing supply and demand. While true supply and demand can be very hard to measure, we can at least get a few broad parameters out there to help us all have a common base of facts. The goal of this post is just to present a few basic factual parameters, and then to advance a different way of thinking about supply and demand.

So, let’s do supply first, how many houses are there?

Source. Note: Datawrapper seems to be insisting on their home-base German-language standard formatting items these days.

Cool! So there are about 135 million housing units in the United States! That’s neat! Now, to be clear, that’s not the number of occupied housing units. Just occupiable housing units total. And while the trend is definitely upwards, it does seem like it’s slowed down a bit recently.

We can also see some growth there, about a 2.3% increase in the national stock of housing units since the 2010 Census. However, it should be noted that, as of 2015, national population has increased by about 4.1% since the Census, so, nationally, housing on average should be “more crowded.” Here’s a chart of housing unit growth rates versus population growth rates from 1990 to 2015:

Overall, since 1990, the housing supply has grown 32%, while population has grown just under 29%. Obviously, when you’ve got more people, you need more houses, unless there’s a change in the number of people living in housing units (i.e. homeless), or the average household size changes. While these changes do occur, barring really astonishing events, they tend to be fairly marginal changes, usually not big enough to totally alter the long-run similarity between housing units and population.


Geography of Supply Tightness

We can also think about how housing and population have changed geographically. The map below shows every county in America, color coded by the % change in population since 2010 minus the % change in housing units. In other words, big population growth accompanied by a big housing boom would yield some middling-grayish color. But big population growth accompanied by housing stagnation or decline yields the dark red color. Big housing boom without population growth would yield the dark blue color. Population decline combined with stable or rising housing would also yield the blue color.

Clever readers will of course respond by saying, “But Lyman, growth in households matters more than growth in population.” Well, true, except that you can’t form new households without new houses, ergo, “household formation,” which presumably is a kind of “demand metric” is itself determined by housing unit availability, to at least some extent.

Does the map above match your preconceptions? Did you think Florida and the Texas metro areas had tight housing supply, i.e. population growth substantially outpacing housing supply? We can definitely see the two coasts, but it also looks like all the Sunbelt cities have “tight supplies.” Whoopsie.

It turns out, housing supply lags housing demand pretty well everywhere that there’s been economic growth.

We can map this another way. We can take the number of people per housing unit for each county in 2010, and assume that new housing should come online at a rate equal to whatever is needed to maintain that equivalent ratio. That tells us how many housing units ought to be added, for a given population change. We can then compare that to how many were added, and thereby get a sense of which counties have the most “missing houses,” or, alternatively, the most “empty houses,” so to speak. Or just empty bedrooms, as the case may be.

This map restricts itself to counties where the “gap” has an absolute value of 10,000 or more, basically because if I try to do all counties Datawrapper freaks out at me, and I worry about provoking our robot overlords too frequently or aggressively.

This map does make the situation a bit clearer, however. While almost every major population center is struggling to provide enough housing to sustain the population at a stable household size (implying that people will need to cram together more tightly), some seem pretty obviously worse than others. Florida and California in particular seem to be struggling to ensure that housing supply keeps up with rising population.

We can also look at how supply tightness relates to local prices. The ACS asks about home values. Sadly the 1-year ACS samples don’t cover even close to all counties, we just have the 807 largest counties for the 1-year gross value data. It turns out, this relationship is not extremely strong across the whole series. Areas where housing supply rose much faster relative to population did not have consistently different price performance. However, some caveats: I’m excluding a huge set of counties that are disproportionately likely to have housing oversupply (more rural counties). Also, ACS’ method does not control for differences in housing characteristics across time or place.

That said, if, among my 807 largest counties, I compare the 100 “tightest” by my metrics to the 100 “least tight,” I do find some striking differences. The average “tightest” supply counties saw home values rise by about 4%. The average “least tight” supply counties saw home values fall by about 0.4%. These differences are statistically significant, with a t-stat of 3.6. In other words, although the relationship between supply and price is not perfectly consistent, it nonetheless does seem that areas with the most abundant housing supply have more moderated prices than areas with the most restricted housing supply.

However, another caveat: the areas where housing supply was very abundant were sometimes also areas of significant economic malaise, which suppresses population growth, thus making it easier to score “very abundant housing supply.” So this is all very, very rough.

But we can do a little bit better. In one state in particular, we have enough counties to basically introduce a state-specific dummy variable: California. Maybe something specific to certain states impacts the cost of living without directly impacting my “tightness” metric. Let’s look at California. Most states we really don’t have enough data points to get a good sample, but, in those we do, we can see some fun stuff.

You won’t be surprised to hear it’s about a 70% correlation. Which means we can explain about 70% of the variation in home value changes in California simply by asking how many were added versus how the local population changed. Very simply supply, very simple demand, most of the answer settled. Getting that other 30%, of course, can be very hard.

We can also look at Florida. For Florida, I dropped one county out of the equation: Sumter Co. That county includes The Villages, and the huge change in local housing stock quality makes a strong appearance there, dramatically raising prices despite fairly little overall supply tightening.

It’s not quite as neat a story, but, still, it’s hard not to see it once you’ve seen it.

Or, we can look at Texas.

Woah, okay. What’s going on here? In Texas, there’s basically no relationship between changes in the tightness of supplies and population. That’s weird. I wonder if there’s something related to housing policy where California is at one extreme, Texas is at the other, and Florida is somewhere between them that could help explain why local supply tightness is almost fully capitalized into local prices in California but not in Texas? I dunno, I can’t think of any elegantly simple explanation for this, can you?

(spoiler: zoning rules and physical space constraints are the commonly-cited drivers of these different relationships. Specifically, rules or geography that restrict greenfield development. They do matter. But I’ve got a 3rd preferred candidate in mind to be revealed below.)


Sources of Housing Supply Volatility

But this whole exercise got me thinking: just how big are shocks to housing supply? I honestly have no idea what share of the stock of structures in the country depreciates beyond usability, is relocated, is opened for habitation, etc, in a given year. That’s a total mystery to me.

So I figured I’d look at what I assume simply has got to be the most volatile housing supply situation in the country in recent memory: New Orleans around Hurricane Katrina. If we group together all the Louisiana counties with lower housing units total in 2006 than 2005, we can get our “Hurricane-impacted area.” Here’s their housing unit chart:

It’s about a 12% decline in housing units after Hurricane Katrina. That’s net, of course; gross housing units lost would be more. This includes new construction and repaired units as they come back online.

So at least on the downside, it’s possible for a major city to have a negative shock in the 10% neighborhood. Not just possible; we’ve seen this happen, within the last decade (or just a hair more — yeah guys, Katrina was over a decade ago). It’s rare, sure, and disastrous usually, but not by any means impossible. And the more locally we define our market, the more volatility seems possible.

What about on the upside?

Well, 10% annual change is virtually unheard of for large counties. But 3–5% is quite possible. Just in 2015, there were 6 large counties (over 100,000 people) where the stock of housing rose by more than 3%. In total, 52 counties have had a year of 3% housing stock growth at least once just since 2010. But even so, upside volatility seems rather less of an issue to supply than downside volatility.

We can get an even clearer idea of housing stock changes from the Components of Invintory Change (CINCH) produced by HUD. CINCH goes back to 1985 but there are many method changes and it’s not queryable, so you have to just go PDF-diving. So I’ll start with just the most recent CINCH report, reflecting 2011–13.

Wowza. They did work. Let’s parse this out.

By far the largest single component is new construction, adding almost 1.2 million units to the existing 132.4 million units present in 2011. All the other “addition” components like conversion from nonresidential units, changes in survey method, repairs, etc, come to just about 820,000 units. So new construction yielded about 0.88% gross increase, and all other sources boosted the measured housing supply by about 0.62%.

Sources of housing unit loss are more diversified. I’ll lump them into two categories: decline in physical quality of housing, or other. The first category is for demolition, disaster, condemned housing, or badly damaged housing. This amounts to about 680,000 lost housing units from 2011–2013, or about 0.5% of the 2011 stock eliminated. Meanwhile, all other sources of loss, like conversion to nonresidential use or housing mergers, for example, amount to 885,000 lost units, or 0.67% of the 2011 supply.

When you add up all the moving parts here, you get about 3.5 million housing units “in flux.” This is your total potential supply shock or re-allocation, so to speak. About 2.7%. Some years it may be more, some less, but my understanding is that 2–3% is considered normal.

This data is cool and it shows us some sources of potential supply volatility. Housing supply can be altered by disasters, by changes in minimum habitability laws, by construction regulations, by mobile-home activity or migration, by changing economic returns to nonresidential use, etc. Housing supply is way, way bigger than new home starts. I particularly like the mobile-home bit, because it’s a case where migration actually creates its own housing.

We can also look at CINCH data over time. I’ve combined many categories here due to changes in category definitions across survey years to try and keep things as consistent as possible.

Wowza has construction declined. So, okay, recent years may overstate the importance of non-construction sources of housing supply, as construction has fallen dramatically. Also, I had to make up a construction number for 1999, because the CINCH number was clearly erroneous, so I subbed in a number derived from housing completion data for the year.

Meanwhile, we can also see how nonresidential conversions have changed.

While once conversion from residential to nonresidential was extremely dominant, now, if the “converted warehouse lofts” weren’t a clue to you already, more real estate is being made residential. That’s an important shift to note.

We can also look at all units constructed or repaired, so all “building,” versus all damaged, destroyed, or condemned, so “breaking.

As you can see, while both “building” and “breaking” have fallen in recent years, the decline in building has been vastly greater. The decline in “breaking” is interesting as well, though: one wonders if declines in new units added could lead to homeowners being proactive in protection and maintenance of existing units?

There’s just one teentsy-weentsy problem with all of this. From 2011 to 2013, CINCH says we moved from 132.4 million to 132.8 million housing units. But the American Housing Survey says we moved from 132.2 million to 133.2 million. AHS suggests the housing stock increased by more than double the amount CINCH suggests. Indeed, throughout these periods, CINCH’s estimate of the housing stock can be off by hundreds of thousands, or even a million or more, total housing units. This seems like a big deal to me personally, but I don’t work in the housing-data space as much as the migration-data space, so maybe there’s a good answer for these differences. Plus, as far as I know, CINCH data is not available locally; that is, I can’t tell you how much of the housing supply change in a given metro was due to conversion from nonresidential, or disasters, or some other factor.

All of this discussion of sources of housing stock volatility is a roundabout way of pushing towards broadening what we think about when we think of supply shocks. About 40% of housing units added to the housing stock in a given year are not new housing units. So if you’re only looking at new housing construction, you’re missing half the supply story.

Well, actually, maybe more than half.


What Is Supply Today?

I want to explore the real nature of housing supply and demand using a hypothetical city. I’ll start out with boring conventional supply and demand curves we’ve all seen, then move to something more fun. So if you’re economically-minded and get SD graphs, you can skip down a bit. For the rest, bear with me, I have an interesting conclusion.

Widget Inc and Migrationville, USA

Say we have a city, called Migrationville of course, with 2.4 million people, and a housing stock of 1 million units. In some sense, the local housing “supply” can be seen as 1 million units, demand as 2.4 million people. We could then go across time and say that, if we assume supply and demand were in equilibrium at 2.4 and 1, if we add 125,000 people and 50,000 more houses, maybe that is the supply and demand we should look at: the marginal increase.

But now think for a second about actual housing markets. Not all houses are for sale at a given time, nor are all people looking to buy. At any given time, a discrete number of buyers and sellers are on the market. Conventionally, we expect that, as prices rise, more sellers will enter the market, but fewer buyers. Put simply, maybe you won’t sell your house for $200,000, but you would consider it for $250,000, and at $300,000 you’re actively looking for buyers. A buyer’s situation is, of course, reversed. Here’s our fun little supply/demand graph (each X-axis unit of 1 can be imagined as being 10,000 houses; I’ll use Excel because putting all these in Datawrapper would be a pain, and you don’t need the “data” since this is all hypothetical):

So when the demand-curve shifts, supply responds with a different price and quantity offered. To this neat, simple model of supply and demand we can then exogenously add more supply. I mentioned 50,000 more houses above; let’s say that we add those 50,000 houses, with no new residents, i.e. no demand change. What happens to price? Well, as the Supply 2 line shows, it falls!

But let’s say that all these new buildings were ugly. The existing homeowners hated those nasty new buildings so much, they forbade construction of anything new. So quantity of housing is fixed at the equilibrium of Demand 1 and Supply 2. We get this graph:

My graphing program doesn’t support making the line jump straight up, so consider that to be a straight jump up. Now imagine that a new company (Widgets, Inc) opens up in Migrationville and employs 25,000 workers. Remember, we’ve prohibited all new housing construction. This new demand curve gives us:

Hot dang! House prices rose from 97 to 124, a huge increase!

But hold on. If there are 25,000 more households in town and nobody built new houses, where are they all living? This makes no sense. If we had a total ban on new housing supply, then these 25,000 households could not have been added at any price. There must have been pure displacement; they just bid locals out of their own homes. But then we have to ask: why were locals selling? What is this strange thing happening where we get changes in the distribution and allocation of supply? What is the actual housing market?

The housing supply graph everyone has in their head is a simple, long-run graph.

That long-run graph is more-or-less an equilibrium graph of new residential construction, which is not the supply graph for housing!

Brace yourselves everybody. We gonna’ get weird here.

The total stock of housing units is not housing supply. Even in an unrestricted market with no zoning, housing supply and demand will not equilibrate at the intersection of total population demand and total housing supply. In the long run, if there were never any shocks to the economy and preferences were perfectly stable, we would see this kind of housing-unit-driven equilibrium. But never in the real world. Births and deaths constantly create exogenous shocks to residency and housing preferences, as do deportations and incarcerations. Plus, preferences do change. And, of course, the transaction costs to housing changes are positively staggering. You can end up paying a year’s salary just in frictional costs if you’re not careful! With such cavernously vast transaction costs and time-inconsistency problems prepared to swallow market participants whole, our basic assumption should be that new housing construction and population growth are at best only indirect measures of supply and demand.

Real markets clear based on the preferences and resources of buyers and sellers with various degrees of stickiness. For example, maybe Migrationville wasn’t doing so well economically before Widget Inc showed up. Maybe there are spillover jobs now, making existing residents less likely to put their houses on the market. The existence of strong demand can be directly associated with reduced supply if that strong demand alters local property owners’ expectations about future prices (or their own future incomes!)! Nobody wants to sell their house when the neighborhood is “on the way up.” You sell when you think it’s about to go down. If this kind of “positive shock to expectations” occurs, then maybe we can model Migrationville’s housing supply like this:

Or, heck, maybe we get weirder. Maybe we decide that expectations about future prices are endogenous to the price-setting function already embedded in the supply curve, so any time prices rise, expectations of future prices also rise. If this is true, then any time there’s a positive demand shock, the supply curve should get steeper. In which case, we wouldn’t shift the supply curve up/down left/right, we would tilt it. Wheeeeeeeee!

In other words, this model presumes that any time prices rise, potential sellers recalibrate their expectations on the assumption that recently observed trends will continue. Is that really what happens? No frickin clue, dude! But the point is that it’s not entirely unreasonable. And the primary determinant of if it happens or not will be the holding cost of housing versus the frictional cost of buying or selling.

If hanging onto your house gets more and more expensive as prices rise around you, there’s a high and rising holding cost of housing. Assuming the amenity that housing yields is unchanged with respect to price shocks (which is not true if price shocks are accompanied by positive income shocks), then local policy and cost regimes that force homeowners to bear the cost of local prices will induce people to sell more in response to rising prices. On the other hand, if homeowners do not face a rising holding cost of housing, they may sit and wait for a better price for longer. The primary market-responsive variable in the holding cost of housing is the property tax bill. When property taxes are abated, or assessments delayed, or anything is done to shield property owners from paying a bill equal to the full, current market value, well then, that will alter supply. In other words, higher property taxes can make housing markets respond to supply/demand disequilibria faster.

You will not be surprised to learn that property taxes in Texas are much higher as a percent of aggregate home values than in Florida, and Florida is higher than California. Surprise! A more rational distribution of the tax burden promotes more efficiently-functioning markets!

Looking at the other side of the equation, factors that alter the transaction cost to buying or selling would seem to matter a great deal. The more paperwork, inspections, or legal fees required to meaningfully enter or exit the market as a buyer or seller, the more likely that people who might like to sell don’t, or who might like to buy can’t. In other words, local rules and norms about realtor commissions and required inspections can suppress supply and/or demand. As policy, technological, or social conditions change, it may alter these frictional costs. The cost of moving is another such frictional cost that may be particularly high for certain classes of homeowners like the elderly.

Changes to the holding cost of housing (i.e. property taxes, variable rate mortgages, demand for AirBnB service, etc) can be seen as straightforward shifts in the supply curve. Changes in the frictional costs associated with market participation are more complex, and could be associated with either the demand or supply curves.

Looking at a totally different set of factors yields other possibilities. Widget Inc might cause local wage costs to rise if they’re a much more capital-intensive firm than other local firms, i.e. if they have a way bigger profile than local companies. So maybe Widget Inc is a multinational relocating huge production facilities. This may cause local firms to have to bid up wages. Any firm buying local labor but selling beyond the locality, like small manufacturing for example, may be put out of business. They will then sell their property, which may reduce nonresidential property values. But use of that property for employer-businesses is no good, so it becomes residential real estate. Remember that component of CINCH? That was over 400,000 housing units worth of churn!

But then again, if local businesses fire workers, maybe that causes an influx of new housing onto the market! Goodness, this does get complicated doesn’t it? And what if Migrationville becomes a leisure destination and you get lots of absentee owners, bidding up the price of housing while reducing the resident population? And what if Migrationville residents are able to turn their homes into short-term hotels, creating additional income? Will that be capitalized into land values, rents, and property taxes?

I won’t graph all of these. But the point is, all of these factors impact supply, even if some of them we might consider to be “demand.” When a firm fires workers, we may see this as a negative demand shock. But if those workers then move away, they’ll try to sell their houses, which puts those houses on the market even though local prices are unchanged or have even fallen. That’s a textbook definition of a supply shock! Ask somebody what “supply” looks like in factory town after a big layoff. Suddenly, market supplies look rather abundant. How often in discussions of housing supply do you hear about the determinants of occupied-housing-sales? Or is it all just construction of new houses?

Lessons from Migrationville

Of course, Migrationville could add more supply! Build, baby, build! And that would help! But we need to keep in mind that new construction is not the dominant component in actual housing markets.

Real estate agencies suggest that existing houses will make up about 5 million units sold in a year, give or take a million, while new homes amount to just 500,000 or so.
The dominant factor in short- and medium-term housing costs is not residential construction, but factors inducing buyers and sellers to enter and exit the market for existing homes.

The same goes, by the way, for renting. The valid metric for rental costs isn’t the average rent paid by all people, but the rent you could obtain if you sought to rent a unit now. So when we think about rental supply/demand, we shouldn’t think about the number of rental units and the population, we should think about how many people are looking to rent right now versus how many people are looking to take on renters right now. If lots of people lose their formerly-owned homes, that could be a positive demand shock to rental units, for example.


Conclusion

My goal here was not to advance some grand theory of housing supply, but to complicate the simple narratives told about housing supply, policy, and prices. It’s not as simple as new housing starts compared to households or something like that. The determinants of the housing stock are numerous and diverse, and the policies impacting actual supply range much more widely than just “zoning” and “land use.” Housing construction is just one factor, in recent years amounting to about 1/3 or so of total gross changes in the housing stock. The housing stock itself is not what actually determines housing market dynamics: actual market listings and purchases are what really matter. Those actual listings and purchases depend on a huge variety of factors, with supply and demand tied together in a way unlike what we see in many product and service markets. Computer companies don’t withhold sales of computers in the hopes that tomorrow’s prices will be better. Homeowners do. Considering the complexities of actual housing markets suggests that determinants of high prices could be due to much more than restrictions on new housing construction. Considering the different incentives facing actual market participants suggests that a focus on zoning and land use may be leaving potentially large allocative efficiencies on the table. And, in case you didn’t notice the quiet build-up through this whole post, surprise, that’s why we should have a Georgist land tax! You just got the economics equivalent of rick-rolled.

Check out my Podcast about the history of American migration.

If you like this post and want to see more research like it, I’d love for you to share it on Twitter or Facebook. Or, just as valuable for me, you can click the recommend button at the bottom of the page. Thanks!

Follow me on Twitter to keep up with what I’m writing and reading. Follow my Medium Collection at In a State of Migration if you want updates when I write new posts. And if you’re writing about migration too, feel free to submit a post to the collection!

I’m a graduate of the George Washington University’s Elliott School with an MA in International Trade and Investment Policy, and an economist at USDA’s Foreign Agricultural Service. I like to learn about migration, the cotton industry, airplanes, trade policy, space, Africa, and faith. I’m married to a kickass Kentucky woman named Ruth.

My posts are not endorsed by and do not in any way represent the opinions of the United States government or any branch, department, agency, or division of it. My writing represents exclusively my own opinions. I did not receive any financial support or remuneration from any party for this research. More’s the pity.